nx_atlas_graphrec_gate.nx

nx_atlas_graphrec_gate.nx -- proves the RECOMBINATION SCORING, not merely that the organ runs. WHY IT EXISTS: I shipped nx_atlas_graphrec LIVE (built, promoted, registered, called over /mcp, committing to the discovery-graph- plane) WITHOUT a gate -- a rule-9 violation of this ecosystem's own standard, self-caught. Worse, the scoring was inline in main(), so the only way to exercise it was to run the whole organ against a real 17k-node graph; any gate over that would have been vacuous. gr_propose was extracted first so these teeth bite the actual decision logic against SYNTHETIC graphs where the right answer is known by construction. Each tooth isolates ONE rule, and the two hub rules are separate because they failed separately: the FIRST live run put nx_syscalls (Ca=14084) into 6 of the top 20 because hubs were suppressed as WITNESS but not as ENDPOINT. T4 and T5 exist so that cannot come back. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0